A public-health chatbot hallucinates outbreak numbers that spread widely online, and no containment or communications steps are defined. What should the auditor recommend?

Defining containment and public-communications steps for AI misinformation incidents emphasizes that hallucinated outbreak numbers spreading online require rapid civic response. The correct recommendation works because public-health GenAI errors can amplify harm at social-media speed; IR must disable or constrain the faulty path, issue corrections, and coordinate messaging with health communications—not wait for bureaucratic reporting cycles. Predefined steps reduce improvisation during trust-critical events.

Requiring documented accuracy testing of health statistics before chatbot release fails to address this incident because it is a pre-deployment control that cannot contain or correct outbreak numbers that have already gone viral—testing earlier doesn't undo a hallucination that is already spreading. Advising residents to double-check critical information through official channels is a reasonable caution, but on its own it does not contain the faulty chatbot path or correct the viral false numbers already circulating—it shifts the burden to residents instead of fixing the source. Implementing detailed logging of chatbot prompts and outputs for audit review fails because logging only preserves a record of what happened; it does nothing to stop the faulty responses, correct the public record, or communicate with residents during the active incident.

Exam caveat: correcting one chatbot utterance is incomplete if caches, screenshots, and third-party reposts keep the false numbers alive without a broader communications plan. Operational check: maintain an AI misinformation runbook (kill switch, corrected advisory, social/web updates, rumor monitoring), assign a communications lead, and drill the public-health chatbot scenario before the next outbreak season.